Harnessing Generative AI for Automated Exam Feedback Generation

Abstract

Summative assessments, particularly written exams, remain the most common method of evaluating performance. Although feedback is essential for learning, students often receive no due to lecturers’ time limitations. To address this issue, we employed a design science research approach to identify objectives and design a system based on an automated feedback system described in the literature and extended with generative AI. The system uses a single structured prompt template that enables consistent text generation, speeds up setup, and minimises issues such as prompting demands, hallucinations, and bias. Bias is reduced by excluding student-specific data, and hallucinations are counteracted by review by the lecturer. Bloom’s taxonomy personalises feedback without disclosing the exam. The system guides lecturers through targeted questions answered in bullet points, which are synthesised into comments aligned with components of learner-centred feedback. Three lecturers trialled the system in three exams, demonstrating its capability to support learner-centred feedback generation.

This document is currently not available here.

Share

COinS